generic-library/opencv
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

added masks support for bitwise operations on GPU

Browse Source
This commit is contained in:
Alexey Spizhevoy
2010-11-22 09:39:34 +00:00
parent 3163cfb845
commit 732bd621bb
4 changed files with 293 additions and 134 deletions
Download Patch File Download Diff File Expand all files Collapse all files

137
modules/gpu/src/arithm.cpp
View File

@@ -81,14 +81,14 @@ void cv::gpu::cartToPolar(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, bool)
void cv::gpu::cartToPolar(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, bool, const Stream&) { throw_nogpu(); }
void cv::gpu::polarToCart(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, bool) { throw_nogpu(); }
void cv::gpu::polarToCart(const GpuMat&, const GpuMat&, GpuMat&, GpuMat&, bool, const Stream&) { throw_nogpu(); }
void cv::gpu::bitwise_not(const GpuMat&, GpuMat&) { throw_nogpu(); }
void cv::gpu::bitwise_not(const GpuMat&, GpuMat&, const Stream& stream) { throw_nogpu(); }
void cv::gpu::bitwise_or(const GpuMat&, const GpuMat&, GpuMat&) { throw_nogpu(); }
void cv::gpu::bitwise_or(const GpuMat&, const GpuMat&, GpuMat&, const Stream& stream) { throw_nogpu(); }
void cv::gpu::bitwise_and(const GpuMat&, const GpuMat&, GpuMat&) { throw_nogpu(); }
void cv::gpu::bitwise_and(const GpuMat&, const GpuMat&, GpuMat&, const Stream& stream) { throw_nogpu(); }
void cv::gpu::bitwise_xor(const GpuMat&, const GpuMat&, GpuMat&) { throw_nogpu(); }
void cv::gpu::bitwise_xor(const GpuMat&, const GpuMat&, GpuMat&, const Stream& stream) { throw_nogpu(); }
void cv::gpu::bitwise_not(const GpuMat&, GpuMat&, const GpuMat&) { throw_nogpu(); }
void cv::gpu::bitwise_not(const GpuMat&, GpuMat&, const GpuMat&, const Stream&) { throw_nogpu(); }
void cv::gpu::bitwise_or(const GpuMat&, const GpuMat&, GpuMat&, const GpuMat&) { throw_nogpu(); }
void cv::gpu::bitwise_or(const GpuMat&, const GpuMat&, GpuMat&, const GpuMat&, const Stream&) { throw_nogpu(); }
void cv::gpu::bitwise_and(const GpuMat&, const GpuMat&, GpuMat&, const GpuMat&) { throw_nogpu(); }
void cv::gpu::bitwise_and(const GpuMat&, const GpuMat&, GpuMat&, const GpuMat&, const Stream&) { throw_nogpu(); }
void cv::gpu::bitwise_xor(const GpuMat&, const GpuMat&, GpuMat&, const GpuMat&) { throw_nogpu(); }
void cv::gpu::bitwise_xor(const GpuMat&, const GpuMat&, GpuMat&, const GpuMat&, const Stream&) { throw_nogpu(); }
cv::gpu::GpuMat cv::gpu::operator ~ (const GpuMat&) { throw_nogpu(); return GpuMat(); }
cv::gpu::GpuMat cv::gpu::operator | (const GpuMat&, const GpuMat&) { throw_nogpu(); return GpuMat(); }
cv::gpu::GpuMat cv::gpu::operator & (const GpuMat&, const GpuMat&) { throw_nogpu(); return GpuMat(); }
@@ -873,10 +873,18 @@ void cv::gpu::polarToCart(const GpuMat& magnitude, const GpuMat& angle, GpuMat&
namespace cv { namespace gpu { namespace mathfunc
{
void bitwise_not_caller(const DevMem2D src, int elemSize, PtrStep dst, cudaStream_t stream);
void bitwise_or_caller(int cols, int rows, const PtrStep src1, const PtrStep src2, int elemSize, PtrStep dst, cudaStream_t stream);
void bitwise_and_caller(int cols, int rows, const PtrStep src1, const PtrStep src2, int elemSize, PtrStep dst, cudaStream_t stream);
void bitwise_xor_caller(int cols, int rows, const PtrStep src1, const PtrStep src2, int elemSize, PtrStep dst, cudaStream_t stream);
void bitwise_not_caller(int rows, int cols, const PtrStep src, int elemSize, PtrStep dst, cudaStream_t stream);
void bitwise_not_caller(int rows, int cols, const PtrStep src, int elemSize, PtrStep dst, const PtrStep mask, cudaStream_t stream);
void bitwise_or_caller(int rows, int cols, const PtrStep src1, const PtrStep src2, int elemSize, PtrStep dst, cudaStream_t stream);
void bitwise_or_caller(int rows, int cols, const PtrStep src1, const PtrStep src2, int elemSize, PtrStep dst, const PtrStep mask, cudaStream_t stream);
void bitwise_and_caller(int rows, int cols, const PtrStep src1, const PtrStep src2, int elemSize, PtrStep dst, cudaStream_t stream);
void bitwise_and_caller(int rows, int cols, const PtrStep src1, const PtrStep src2, int elemSize, PtrStep dst, const PtrStep mask, cudaStream_t stream);
void bitwise_xor_caller(int rows, int cols, const PtrStep src1, const PtrStep src2, int elemSize, PtrStep dst, cudaStream_t stream);
void bitwise_xor_caller(int rows, int cols, const PtrStep src1, const PtrStep src2, int elemSize, PtrStep dst, const PtrStep mask, cudaStream_t stream);
template <int opid, typename Mask>
void bitwise_bin_op(int rows, int cols, const PtrStep src1, const PtrStep src2, PtrStep dst, int elem_size, Mask mask, cudaStream_t stream);
}}}
namespace
@@ -884,75 +892,126 @@ namespace
void bitwise_not_caller(const GpuMat& src, GpuMat& dst, cudaStream_t stream)
{
dst.create(src.size(), src.type());
mathfunc::bitwise_not_caller(src, src.elemSize(), dst, stream);
mathfunc::bitwise_not_caller(src.rows, src.cols, src, src.elemSize(), dst, stream);
}
void bitwise_not_caller(const GpuMat& src, GpuMat& dst, const GpuMat& mask, cudaStream_t stream)
{
CV_Assert(mask.type() == CV_8U && mask.size() == src.size());
dst.create(src.size(), src.type());
mathfunc::bitwise_not_caller(src.rows, src.cols, src, src.elemSize(), dst, mask, stream);
}
void bitwise_or_caller(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, cudaStream_t stream)
{
CV_Assert(src1.size() == src2.size());
CV_Assert(src1.type() == src2.type());
CV_Assert(src1.size() == src2.size() && src1.type() == src2.type());
dst.create(src1.size(), src1.type());
mathfunc::bitwise_or_caller(dst.cols, dst.rows, src1, src2, dst.elemSize(), dst, stream);
mathfunc::bitwise_or_caller(dst.rows, dst.cols, src1, src2, dst.elemSize(), dst, stream);
}
void bitwise_or_caller(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, cudaStream_t stream)
{
CV_Assert(src1.size() == src2.size() && src1.type() == src2.type());
CV_Assert(mask.type() == CV_8U && mask.size() == src1.size());
dst.create(src1.size(), src1.type());
mathfunc::bitwise_or_caller(dst.rows, dst.cols, src1, src2, dst.elemSize(), dst, mask, stream);
}
void bitwise_and_caller(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, cudaStream_t stream)
{
CV_Assert(src1.size() == src2.size());
CV_Assert(src1.type() == src2.type());
CV_Assert(src1.size() == src2.size() && src1.type() == src2.type());
dst.create(src1.size(), src1.type());
mathfunc::bitwise_and_caller(dst.cols, dst.rows, src1, src2, dst.elemSize(), dst, stream);
mathfunc::bitwise_and_caller(dst.rows, dst.cols, src1, src2, dst.elemSize(), dst, stream);
}
void bitwise_and_caller(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, cudaStream_t stream)
{
CV_Assert(src1.size() == src2.size() && src1.type() == src2.type());
CV_Assert(mask.type() == CV_8U && mask.size() == src1.size());
dst.create(src1.size(), src1.type());
mathfunc::bitwise_and_caller(dst.rows, dst.cols, src1, src2, dst.elemSize(), dst, mask, stream);
}
void bitwise_xor_caller(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, cudaStream_t stream)
{
CV_Assert(src1.size() == src2.size());
CV_Assert(src1.type() == src2.type());
dst.create(src1.size(), src1.type());
mathfunc::bitwise_xor_caller(dst.cols, dst.rows, src1, src2, dst.elemSize(), dst, stream);
mathfunc::bitwise_xor_caller(dst.rows, dst.cols, src1, src2, dst.elemSize(), dst, stream);
}
void bitwise_xor_caller(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, cudaStream_t stream)
{
CV_Assert(src1.size() == src2.size() && src1.type() == src2.type());
CV_Assert(mask.type() == CV_8U && mask.size() == src1.size());
dst.create(src1.size(), src1.type());
mathfunc::bitwise_xor_caller(dst.rows, dst.cols, src1, src2, dst.elemSize(), dst, mask, stream);
}
}
void cv::gpu::bitwise_not(const GpuMat& src, GpuMat& dst)
void cv::gpu::bitwise_not(const GpuMat& src, GpuMat& dst, const GpuMat& mask)
{
::bitwise_not_caller(src, dst, 0);
if (mask.empty())
::bitwise_not_caller(src, dst, 0);
else
::bitwise_not_caller(src, dst, mask, 0);
}
void cv::gpu::bitwise_not(const GpuMat& src, GpuMat& dst, const Stream& stream)
void cv::gpu::bitwise_not(const GpuMat& src, GpuMat& dst, const GpuMat& mask, const Stream& stream)
{
::bitwise_not_caller(src, dst, StreamAccessor::getStream(stream));
if (mask.empty())
::bitwise_not_caller(src, dst, StreamAccessor::getStream(stream));
else
::bitwise_not_caller(src, dst, mask, StreamAccessor::getStream(stream));
}
void cv::gpu::bitwise_or(const GpuMat& src1, const GpuMat& src2, GpuMat& dst)
void cv::gpu::bitwise_or(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask)
{
::bitwise_or_caller(src1, src2, dst, 0);
if (mask.empty())
::bitwise_or_caller(src1, src2, dst, 0);
else
::bitwise_or_caller(src1, src2, dst, mask, 0);
}
void cv::gpu::bitwise_or(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const Stream& stream)
void cv::gpu::bitwise_or(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, const Stream& stream)
{
::bitwise_or_caller(src1, src2, dst, StreamAccessor::getStream(stream));
if (mask.empty())
::bitwise_or_caller(src1, src2, dst, StreamAccessor::getStream(stream));
else
::bitwise_or_caller(src1, src2, dst, mask, StreamAccessor::getStream(stream));
}
void cv::gpu::bitwise_and(const GpuMat& src1, const GpuMat& src2, GpuMat& dst)
void cv::gpu::bitwise_and(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask)
{
::bitwise_and_caller(src1, src2, dst, 0);
if (mask.empty())
::bitwise_and_caller(src1, src2, dst, 0);
else
::bitwise_and_caller(src1, src2, dst, mask, 0);
}
void cv::gpu::bitwise_and(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const Stream& stream)
void cv::gpu::bitwise_and(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, const Stream& stream)
{
::bitwise_and_caller(src1, src2, dst, StreamAccessor::getStream(stream));
if (mask.empty())
::bitwise_and_caller(src1, src2, dst, StreamAccessor::getStream(stream));
else
::bitwise_and_caller(src1, src2, dst, mask, StreamAccessor::getStream(stream));
}
void cv::gpu::bitwise_xor(const GpuMat& src1, const GpuMat& src2, GpuMat& dst)
void cv::gpu::bitwise_xor(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask)
{
::bitwise_xor_caller(src1, src2, dst, 0);
if (mask.empty())
::bitwise_xor_caller(src1, src2, dst, 0);
else
::bitwise_xor_caller(src1, src2, dst, mask, 0);
}
void cv::gpu::bitwise_xor(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const Stream& stream)
void cv::gpu::bitwise_xor(const GpuMat& src1, const GpuMat& src2, GpuMat& dst, const GpuMat& mask, const Stream& stream)
{
::bitwise_xor_caller(src1, src2, dst, StreamAccessor::getStream(stream));
if (mask.empty())
::bitwise_xor_caller(src1, src2, dst, StreamAccessor::getStream(stream));
else
::bitwise_xor_caller(src1, src2, dst, mask, StreamAccessor::getStream(stream));
}
cv::gpu::GpuMat cv::gpu::operator ~ (const GpuMat& src)

178
modules/gpu/src/cuda/mathfunc.cu
View File

@@ -243,100 +243,154 @@ namespace cv { namespace gpu { namespace mathfunc
//////////////////////////////////////////////////////////////////////////////
// Per-element bit-wise logical matrix operations
struct Mask8U
{
explicit Mask8U(PtrStep mask): mask(mask) {}
__device__ bool operator()(int y, int x) { return mask.ptr(y)[x]; }
PtrStep mask;
};
struct MaskTrue { __device__ bool operator()(int y, int x) { return true; } };
__global__ void bitwise_not_kernel(int cols, int rows, const PtrStep src, PtrStep dst)
// Unary operations
enum { UN_OP_NOT };
template <typename T, int opid>
struct UnOp { __device__ T operator()(T lhs, T rhs); };
template <typename T>
struct UnOp<T, UN_OP_NOT>{ __device__ T operator()(T x) { return ~x; } };
template <typename T, int cn, typename UnOp, typename Mask>
__global__ void bitwise_un_op(int rows, int cols, const PtrStep src, PtrStep dst, UnOp op, Mask mask)
{
const int x = blockDim.x * blockIdx.x + threadIdx.x;
const int y = blockDim.y * blockIdx.y + threadIdx.y;
if (x < cols && y < rows)
if (x < cols && y < rows && mask(y, x))
{
dst.ptr(y)[x] = ~src.ptr(y)[x];
T* dsty = (T*)dst.ptr(y);
const T* srcy = (const T*)src.ptr(y);
#pragma unroll
for (int i = 0; i < cn; ++i)
dsty[cn * x + i] = op(srcy[cn * x + i]);
}
}
void bitwise_not_caller(const DevMem2D src, int elemSize, PtrStep dst, cudaStream_t stream)
template <int opid, typename Mask>
void bitwise_un_op(int rows, int cols, const PtrStep src, PtrStep dst, int elem_size, Mask mask, cudaStream_t stream)
{
dim3 threads(16, 16, 1);
dim3 grid(divUp(src.cols * elemSize, threads.x), divUp(src.rows, threads.y), 1);
bitwise_not_kernel<<<grid, threads, 0, stream>>>(src.cols * elemSize, src.rows, src, dst);
if (stream == 0)
cudaSafeCall(cudaThreadSynchronize());
dim3 threads(16, 16);
dim3 grid(divUp(cols, threads.x), divUp(rows, threads.y));
switch (elem_size)
{
case 1: bitwise_un_op<unsigned char, 1><<<grid, threads>>>(rows, cols, src, dst, UnOp<unsigned char, opid>(), mask); break;
case 2: bitwise_un_op<unsigned short, 1><<<grid, threads>>>(rows, cols, src, dst, UnOp<unsigned short, opid>(), mask); break;
case 3: bitwise_un_op<unsigned char, 3><<<grid, threads>>>(rows, cols, src, dst, UnOp<unsigned char, opid>(), mask); break;
case 4: bitwise_un_op<unsigned int, 1><<<grid, threads>>>(rows, cols, src, dst, UnOp<unsigned int, opid>(), mask); break;
case 6: bitwise_un_op<unsigned short, 3><<<grid, threads>>>(rows, cols, src, dst, UnOp<unsigned short, opid>(), mask); break;
case 8: bitwise_un_op<unsigned int, 2><<<grid, threads>>>(rows, cols, src, dst, UnOp<unsigned int, opid>(), mask); break;
case 12: bitwise_un_op<unsigned int, 3><<<grid, threads>>>(rows, cols, src, dst, UnOp<unsigned int, opid>(), mask); break;
case 16: bitwise_un_op<unsigned int, 4><<<grid, threads>>>(rows, cols, src, dst, UnOp<unsigned int, opid>(), mask); break;
case 24: bitwise_un_op<unsigned int, 6><<<grid, threads>>>(rows, cols, src, dst, UnOp<unsigned int, opid>(), mask); break;
case 32: bitwise_un_op<unsigned int, 8><<<grid, threads>>>(rows, cols, src, dst, UnOp<unsigned int, opid>(), mask); break;
}
if (stream == 0) cudaSafeCall(cudaThreadSynchronize());
}
void bitwise_not_caller(int rows, int cols,const PtrStep src, int elem_size, PtrStep dst, cudaStream_t stream)
{
bitwise_un_op<UN_OP_NOT>(rows, cols, src, dst, elem_size, MaskTrue(), stream);
}
__global__ void bitwise_or_kernel(int cols, int rows, const PtrStep src1, const PtrStep src2, PtrStep dst)
void bitwise_not_caller(int rows, int cols,const PtrStep src, int elem_size, PtrStep dst, const PtrStep mask, cudaStream_t stream)
{
bitwise_un_op<UN_OP_NOT>(rows, cols, src, dst, elem_size, Mask8U(mask), stream);
}
// Binary operations
enum { BIN_OP_OR, BIN_OP_AND, BIN_OP_XOR };
template <typename T, int opid>
struct BinOp { __device__ T operator()(T lhs, T rhs); };
template <typename T>
struct BinOp<T, BIN_OP_OR>{ __device__ T operator()(T lhs, T rhs) { return lhs | rhs; } };
template <typename T>
struct BinOp<T, BIN_OP_AND>{ __device__ T operator()(T lhs, T rhs) { return lhs & rhs; } };
template <typename T>
struct BinOp<T, BIN_OP_XOR>{ __device__ T operator()(T lhs, T rhs) { return lhs ^ rhs; } };
template <typename T, int cn, typename BinOp, typename Mask>
__global__ void bitwise_bin_op(int rows, int cols, const PtrStep src1, const PtrStep src2, PtrStep dst, BinOp op, Mask mask)
{
const int x = blockDim.x * blockIdx.x + threadIdx.x;
const int y = blockDim.y * blockIdx.y + threadIdx.y;
if (x < cols && y < rows)
if (x < cols && y < rows && mask(y, x))
{
dst.ptr(y)[x] = src1.ptr(y)[x] | src2.ptr(y)[x];
T* dsty = (T*)dst.ptr(y);
const T* src1y = (const T*)src1.ptr(y);
const T* src2y = (const T*)src2.ptr(y);
#pragma unroll
for (int i = 0; i < cn; ++i)
dsty[cn * x + i] = op(src1y[cn * x + i], src2y[cn * x + i]);
}
}
void bitwise_or_caller(int cols, int rows, const PtrStep src1, const PtrStep src2, int elemSize, PtrStep dst, cudaStream_t stream)
template <int opid, typename Mask>
void bitwise_bin_op(int rows, int cols, const PtrStep src1, const PtrStep src2, PtrStep dst, int elem_size, Mask mask, cudaStream_t stream)
{
dim3 threads(16, 16, 1);
dim3 grid(divUp(cols * elemSize, threads.x), divUp(rows, threads.y), 1);
bitwise_or_kernel<<<grid, threads, 0, stream>>>(cols * elemSize, rows, src1, src2, dst);
if (stream == 0)
cudaSafeCall(cudaThreadSynchronize());
}
__global__ void bitwise_and_kernel(int cols, int rows, const PtrStep src1, const PtrStep src2, PtrStep dst)
{
const int x = blockDim.x * blockIdx.x + threadIdx.x;
const int y = blockDim.y * blockIdx.y + threadIdx.y;
if (x < cols && y < rows)
dim3 threads(16, 16);
dim3 grid(divUp(cols, threads.x), divUp(rows, threads.y));
switch (elem_size)
{
dst.ptr(y)[x] = src1.ptr(y)[x] & src2.ptr(y)[x];
case 1: bitwise_bin_op<unsigned char, 1><<<grid, threads>>>(rows, cols, src1, src2, dst, BinOp<unsigned char, opid>(), mask); break;
case 2: bitwise_bin_op<unsigned short, 1><<<grid, threads>>>(rows, cols, src1, src2, dst, BinOp<unsigned short, opid>(), mask); break;
case 3: bitwise_bin_op<unsigned char, 3><<<grid, threads>>>(rows, cols, src1, src2, dst, BinOp<unsigned char, opid>(), mask); break;
case 4: bitwise_bin_op<unsigned int, 1><<<grid, threads>>>(rows, cols, src1, src2, dst, BinOp<unsigned int, opid>(), mask); break;
case 6: bitwise_bin_op<unsigned short, 3><<<grid, threads>>>(rows, cols, src1, src2, dst, BinOp<unsigned short, opid>(), mask); break;
case 8: bitwise_bin_op<unsigned int, 2><<<grid, threads>>>(rows, cols, src1, src2, dst, BinOp<unsigned int, opid>(), mask); break;
case 12: bitwise_bin_op<unsigned int, 3><<<grid, threads>>>(rows, cols, src1, src2, dst, BinOp<unsigned int, opid>(), mask); break;
case 16: bitwise_bin_op<unsigned int, 4><<<grid, threads>>>(rows, cols, src1, src2, dst, BinOp<unsigned int, opid>(), mask); break;
case 24: bitwise_bin_op<unsigned int, 6><<<grid, threads>>>(rows, cols, src1, src2, dst, BinOp<unsigned int, opid>(), mask); break;
case 32: bitwise_bin_op<unsigned int, 8><<<grid, threads>>>(rows, cols, src1, src2, dst, BinOp<unsigned int, opid>(), mask); break;
}
if (stream == 0) cudaSafeCall(cudaThreadSynchronize());
}
void bitwise_and_caller(int cols, int rows, const PtrStep src1, const PtrStep src2, int elemSize, PtrStep dst, cudaStream_t stream)
void bitwise_or_caller(int rows, int cols, const PtrStep src1, const PtrStep src2, int elem_size, PtrStep dst, cudaStream_t stream)
{
dim3 threads(16, 16, 1);
dim3 grid(divUp(cols * elemSize, threads.x), divUp(rows, threads.y), 1);
bitwise_and_kernel<<<grid, threads, 0, stream>>>(cols * elemSize, rows, src1, src2, dst);
if (stream == 0)
cudaSafeCall(cudaThreadSynchronize());
bitwise_bin_op<BIN_OP_OR>(rows, cols, src1, src2, dst, elem_size, MaskTrue(), stream);
}
__global__ void bitwise_xor_kernel(int cols, int rows, const PtrStep src1, const PtrStep src2, PtrStep dst)
void bitwise_or_caller(int rows, int cols, const PtrStep src1, const PtrStep src2, int elem_size, PtrStep dst, const PtrStep mask, cudaStream_t stream)
{
const int x = blockDim.x * blockIdx.x + threadIdx.x;
const int y = blockDim.y * blockIdx.y + threadIdx.y;
if (x < cols && y < rows)
{
dst.ptr(y)[x] = src1.ptr(y)[x] ^ src2.ptr(y)[x];
}
bitwise_bin_op<BIN_OP_OR>(rows, cols, src1, src2, dst, elem_size, Mask8U(mask), stream);
}
void bitwise_xor_caller(int cols, int rows, const PtrStep src1, const PtrStep src2, int elemSize, PtrStep dst, cudaStream_t stream)
void bitwise_and_caller(int rows, int cols, const PtrStep src1, const PtrStep src2, int elem_size, PtrStep dst, cudaStream_t stream)
{
dim3 threads(16, 16, 1);
dim3 grid(divUp(cols * elemSize, threads.x), divUp(rows, threads.y), 1);
bitwise_xor_kernel<<<grid, threads, 0, stream>>>(cols * elemSize, rows, src1, src2, dst);
if (stream == 0)
cudaSafeCall(cudaThreadSynchronize());
bitwise_bin_op<BIN_OP_AND>(rows, cols, src1, src2, dst, elem_size, MaskTrue(), stream);
}
void bitwise_and_caller(int rows, int cols, const PtrStep src1, const PtrStep src2, int elem_size, PtrStep dst, const PtrStep mask, cudaStream_t stream)
{
bitwise_bin_op<BIN_OP_AND>(rows, cols, src1, src2, dst, elem_size, Mask8U(mask), stream);
}
void bitwise_xor_caller(int rows, int cols, const PtrStep src1, const PtrStep src2, int elem_size, PtrStep dst, cudaStream_t stream)
{
bitwise_bin_op<BIN_OP_XOR>(rows, cols, src1, src2, dst, elem_size, MaskTrue(), stream);
}
void bitwise_xor_caller(int rows, int cols, const PtrStep src1, const PtrStep src2, int elem_size, PtrStep dst, const PtrStep mask, cudaStream_t stream)
{
bitwise_bin_op<BIN_OP_XOR>(rows, cols, src1, src2, dst, elem_size, Mask8U(mask), stream);
}
}}}